Buyer shortlist questions for AI search agencies
Choose an AI search provider by what you need done: measurement, diagnosis or implementation. Compare their evidence and scope before comparing labels.
Founder of TomHow, based in Stevenage. SEO and AI search consultancy.
How should you choose an AI search provider?
Choose an AI search provider by the work you need done and the evidence they can show. A tracking tool records answers; an audit diagnoses the gaps; managed support implements changes. Ask which platforms and modes are checked, whether exact questions and source URLs are preserved, what is included in delivery, and who reviews the findings. A visibility score or a single screenshot does not establish that a provider can improve your business.
A four-step shortlist map that helps a UK B2B buyer separate AI search agencies, GEO tools and SEO firms, then route to a dated audit.
Name the job before choosing the provider
If you cannot tell how your business appears in selected answers, start with measurement. If you have observations but no explanation or priorities, diagnosis is the missing job. If the priorities are clear but nobody can carry them out, you need implementation capacity.
For example, a team that already has a developer and an editor may need a focused audit rather than another retainer. A team without those people should check who will actually change the pages before buying more tracking.
Ask to inspect one complete observation
Request the question, answer platform, mode, capture date, returned answer and linked sources from the same observation. That lets you check what was actually measured. A presentation that shows only a score leaves the underlying result unclear.
Ask whether the observation came from an API or the interface customers use. Those are different environments. A provider should explain its coverage and limits without presenting one as a complete record of the other.
Compare the deliverables and exclusions
Put the proposed scopes side by side: question panel, platform coverage, technical checks, page review, changes included, reporting frequency and the person accountable for delivery. A diagnostic report and a monthly implementation service are different purchases.
Check access requirements, software costs and what happens when a platform is unavailable. Confirm whether you receive an export and can use the findings with your own team if you decide against continuing.
Treat promises as a reason to ask better questions
Ask what would count as a useful change and what could not be concluded from it. A citation in one answer is not proof of a recommendation, a lead or a sale. Results from a small panel should be described with the panel and conditions attached.
Google treats generative search work as part of SEO. Be cautious when a pitch depends on a special AI file, an arbitrary content formula or guaranteed inclusion rather than useful pages and a verifiable scope.
Check the platform guidance.
Continue with the useful next step.
Thomas Howcroft
Founder of TomHow and senior operator for SEO, GEO, AEO, and AI search evidence.
View LinkedIn profile ->